Classification of Ciprofloxacin Tablets Using Near-Infrared Spectroscopy and Chemometric Modeling. Issue 8 (August 2017)

• Record Type: Journal Article
• Authors: Fuenffinger, Nathan; Arzhantsev, Sergey; Gryniewicz-Ruzicka, Connie
• Is Part Of: Applied spectroscopy. Volume 71:Issue 8(2017)
• Page Start: 1927
• Page End: 1937
• Publication Date: 2017-08
• Subjects: Ciprofloxacin -- pharmaceutics -- near-infrared spectroscopy -- NIR -- chemometrics -- classification; Spectrum analysis -- Periodicals; 543.505
• DOI: 10.1177/0003702817699624
• Languages: English
• ISSNs: 0003-7028

Abstract

Exposure to unknown, mislabeled, and counterfeit pharmaceutical products is a worldwide problem that presents a serious risk to public health. Near-infrared (NIR) spectroscopy can serve as a useful tool for screening pharmaceuticals in a rapid and cost-effective manner to ensure that drug products are safe and effective. By applying chemometric techniques to NIR spectra from finished products in tablet form, minor spectral differences are discoverable, even in instances where the tablets being evaluated contain the same active pharmaceutical ingredients (APIs). Differences in NIR spectra can occur as a result of various factors including the types and quantities of pharmaceutical excipients used to generate the product and associated manufacturing site process variables. In this study, variability in the NIR spectra of intact tablets with the same API was evaluated using an unsupervised chemometric technique in the form of hierarchical cluster analysis (HCA) on a data set consisting of NIR spectra from more than 800 ciprofloxacin tablets from six manufacturers. Results obtained from HCA and squared Euclidean distance measurements indicate the largest dissimilarities in NIR spectra occur between manufacturers. Based on these findings, a quadratic discriminant analysis (QDA) model was built following dimensionality reduction by principal component analysis for the purpose of predicting the origin of ciprofloxacin tablets. Using QDA, we were able to correctly classify a collection of 907 tablets with greater than 96% accuracy. Chemometric models such as the one developed here could ultimately be employed as part of a large, diversified drug surveillance program.