Application of Multivariate Strategies to the Classification of Pharmaceutical Excipient Manufacturers Based on Near-Infrared (NIR) Spectra

Abstract

Using partial least square discriminate analysis (PLSDA), we studied the spectroscopic differences between the commonly used filler-binder microcrystalline cellulose (MCC) from five manufactures. These samples had subtle differences in the chemical and physical properties, which are often the cause of differences in excipient performance. Studying these differences allowed us to build and validate a model to classify five manufacturers of MCC using near-infrared (NIR) spectra. The sample training set includes 39 MCC samples collected from five manufactures with regions spanning the United States of America, Japan, Taiwan, Germany, and Brazil. The samples from individual manufacturers include diverse grades that differ in moisture content, particle size, and bulk density. Optimized pretreatment methods were identified as standard normal variate normalization, followed by Savitzky-Golay second derivative, mean centering, and orthogonal signal correction. The model was optimized with cross-validation and validated with an independent sample set comprising nine samples collected from those five manufacturers. The results showed that none of the samples in the independent validation set was misclassified. The score and loading plots revealed that the differences in content of oxidized cellulose group, water content and states, hydrogen bonding, and degree of polymerization of the MCC samples are responsible for the class differentiation. Permutation test demonstrated that the outcome of the PLSDA model was significantly different from that of the randomly generated model. The advantages and limitations of the method in this type of application were discussed.