Near infrared spectroscopy (NIRS) was utilized to determine the water content during the drying of a drug substance with Karl Fisher titration as the reference measurement. NIRS calibration models were built with Partial Least Squares (PLS) regression based on spectral region of 1822-1948 nm for samples with 1-40% (w/w) water from five batches. A standard error of prediction (SEP) of 1.85% (w/w) was obtained from validation of the model with additional batches. A second NIRS calibration model using PLS was constructed in the region of 1-10% (w/w) water with samples from the same five calibration batches. This calibration model improved the accuracy of the prediction in the critical region around the end point of drying and provided a standard error of prediction 0.42% (w/w). In addition, direct spectral analyses with Principal Component Analysis (PCA) and peak ratios were applied to distinguish between surface (unbound) water and bound water incorporated into the crystal lattice of the drug substance. Direct spectral analysis indicated the existence of significant numerical and graphical differences between samples containing both surface and bound water, and those containing bound water only. Applying this method to monitor an actual drying process with the graphical assistance of spectral analysis, the drying process can be controlled, and the end point of drying clearly determined to ensure the desired hydrate form of the product. This study has demonstrated the in-line monitoring capability of NIRS to differentiate the surface and bound water simultaneously in addition to the determination of the water level.
Copyright 2003 Wiley-Liss, Inc. and the American Pharmaceutical Association J Pharm Sci 92:1058-1065, 2003