Prospects for multivariate classification of a pharmaceutical intermediate with near-infrared spectroscopy as a...

NIR spectroscopy was applied to develop a fast and reliable quality control system for a pharmaceutical substance to support information obtained through PAT surveillance of its manufacturing process. After calculating different quantitative calibrations of the substance's key quality parameters, a general classification model has been derived to capture the over-all product grade. The final spectral quality conformity model consisting of 96 representative batches - covering high process variability - was sensibilized toward five important quality parameters by their incorporation as PLS responses. The model characteristics were extensively investigated and interpreted to derive a reasonable limit for the reduced chemometric summary quality measure (Hotteling's T(2)). Through this parameter new batches can be assessed easily by their NIR spectra, using versatile test batches for confirmation. Different sets of good quality batches, bad production batches beyond the respective chemical quality limit and synthetic batches exactly at the limit could be accurately assigned through their multivariate evaluation to a large extend. However, high model sensitivity to non-relevant product properties can lead to limited applicability of the model. This may be caused by restricted bandwidth of quality parameters in production environment for calibration, repack effects and high process instability.