Quantitative solid state nuclear magnetic resonance of mixtures utilizing chemometric techniques

Quantitative analysis of solid-state mixtures is essential to the world of pharmaceutical dosage forms. Tablets are often the primary mode of drug delivery. Several techniques can be used to quantify the components within a tablet; advantages and disadvantages of each are discussed. In this paper a new method utilizing solid-state Nuclear Magnetic Resonance spectroscopy (ssNMR) for the absolute quantification of pharmaceutical-like mixtures is proposed. The goal of the proposed method is to utilize chemometrics to build models for the prediction of concentrations of components in mixtures with a high degree of accuracy and precision. Preliminary experimental and theoretical studies were performed to characterize limitations of the instrument and the proposed chemometrics techniques.;The culmination of this research was an extensive study on the factors that affect the prediction of concentration of complex mixtures. Data were collected in a manner which allowed for many models to be created in a systematic manner. Model-based predictions were statistically compared utilizing the T and F Tests. The factors which did not affect the predictive ability of the models include: region of the spectra, binning, line broadening, and the use of data from one probe tune vs. multiple tuning experiments on the same sample. Varying these factors does not add value and therefore can be eliminated from models which results in less time to acquire data and complete the analysis. The factors that had an effect on the predictive ability of the model are: signal to noise ratio and Bloch decay vs. cross polarization. Cross polarization experiments are the most time efficient method of collecting data for Partial Least Squares (PLS) analysis.;The goal of this study has been achieved through the development of a new method for making reliable quantitative predictions. The mean error of prediction for this method is generally less than 2 mg for a sample with a total mass of 250 mg, or less than 1% absolute. The best model was constructed with the following characteristics: cross polarization data, largest signal to noise ratio, 20 Hz line broadening, no binning of data, utilization of the entire spectra acquired, and the use of automatic phasing.