The following is an abstract for the selected article. A PDF download of the full text of this article is available here. Members may download full texts at no charge. Non-members may be charged a small fee for certain articles.

Discerning Some Tylenol Brands Using Attenuated Total Reflection Fourier Transform Infrared Data and Multivariate Analysis Techniques

Volume 64, Number 6 (June 2010) Page 657-668

Msimanga, Huggins Z.; Ollis, Robert J.

Principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) were used to classify acetaminophen-containing medicines using their attenuated total reflection Fourier transform infrared (ATR-FT-IR) spectra. Four formulations of Tylenol (Arthritis Pain Relief, Extra Strength Pain Relief, 8 Hour Pain Relief, and Extra Strength Pain Relief Rapid Release) along with 98% pure acetaminophen were selected for this study because of the similarity of their spectral features, with correlation coefficients ranging from 0.9857 to 0.9988. Before acquiring spectra for the predictor matrix, the effects on spectral precision with respect to sample particle size (determined by sieve size opening), force gauge of the ATR accessory, sample reloading, and between-tablet variation were examined. Spectra were baseline corrected and normalized to unity before multivariate analysis. Analysis of variance (ANOVA) was used to study spectral precision. The large particles (35 mesh) showed large variance between spectra, while fine particles (120 mesh) indicated good spectral precision based on the F-test. Force gauge setting did not significantly affect precision. Sample reloading using the fine particle size and a constant force gauge setting of 50 units also did not compromise precision. Based on these observations, data acquisition for the predictor matrix was carried out with the fine particles (sieve size opening of 120 mesh) at a constant force gauge setting of 50 units. After removing outliers, PCA successfully classified the five samples in the first and second components, accounting for 45.0% and 24.5% of the variances, respectively. The four-component PLS-DA model (R2=0.925 and Q2=0.906) gave good test spectra predictions with an overall average of 0.961 ± 7.1% RSD versus the expected 1.0 prediction for the 20 test spectra used.