Search

The successful implementation of near-infrared spectroscopy (NIRS) in process control of powder blending requires constructing an inclusive spectral database that reflects the anticipated voluntary or involuntary changes in processing conditions, thereby minimizing bias in prediction of blending behavior. In this study, experimental design was utilized as an efficient way of generating blend experiments conducted under varying processing conditions such as humidity, blender speed and component concentration. NIR spectral data, collected from different blending experiments, was used to build qualitative models for prediction of blend homogeneity. Two pattern recognition algorithms: Soft Independent Modeling of Class Analogies (SIMCA) and Principal Component Modified Bootstrap Error-adjusted Single-sample Technique (PC-MBEST) were evaluated for qualitative analysis of NIR blending data. Optimization of NIR models, for the two algorithms, was achieved by proper selection of spectral processing, and training set samples. The models developed were successful in predicting blend homogeneity of independent blend samples under different processing conditions. © 2005 Wiley-Liss, Inc. and the American Pharmacists Association J Pharm Sci 95:407–421, 2006 [ABSTRACT FROM AUTHOR]

The study is aimed at developing a near-infrared (NIR) method for predicting solid fraction (SF) of dry granulated ribbons manufactured with formulation variability. The study investigated the impact of unmodeled chemical variability and regression approaches on method performance. The study utilized an excipient-only formulation system. Calibration compacts were created with chemical and processing variability; followed by collection of NIR spectra. Partial least squares (PLS) and spectral slope algorithms were utilized to model compact SF. Later, the models were deployed to predict SF of test ribbons and compacts containing an API at various concentrations. The risk associated with unmodeled chemical variation manifested itself through generation of new peaks and decreased baseline absorbance in the NIR spectra. The spectral slope was able to better manage this risk, as demonstrated by relatively higher robustness to the increasing load of the active pharmaceutical ingredient (API). The reduced robustness of the PLS approach was attributed to the impact of chemical variability on both spectral baseline and peak absorbance. A prediction error of approximately 5% was observed at 10% drug load using the spectral slope approach. An understanding of the risk associated with unmodeled variability will enable NIR method development as an API sparing technique for low-dose product development. [ABSTRACT FROM AUTHOR]