Deconvolution of chemical and physical information from intact tablets NIR spectra: two- and three-way multivariate...

A three-way calibration strategy has been used to develop a simple method for drug quantitation in intact pharmaceutical tablets. The experimental design defined for the three-way calibration gathers a large variation in drug concentration (15-85% w/w) and compaction pressure (100-500 MPa). These could be process variable ranges during the development of a pharmaceutical formulation following the quality by design (QbD) approach. When a large variation in both chemical and physical variables is defined, the traditional two-way calibration strategies, such as partial least squares (PLS), do not allow the obtaining of multivariate model with good predictive ability (nonlinear predictions and prediction error over 8% w/w). The presented strategy uses parallel factor analysis (PARAFAC) to deconvolute the spectra in scores associated with drug concentration variation, and loadings related with wavelength range and compaction range. The PARAFAC deconvolution was followed by multiple linear regression (MLR) to obtain the simple calibration model with a better predictive ability on the whole compaction pressure range (prediction error less than 1.4% w/w). The proposed three-way strategy can be easily designed and executed, obtaining a multivariate analytical method more robust than using a traditional two-way modeling technique.