Integration of Multivariate Simulation and Method Understanding for Efficient, Science-Based Process Analytical Systems Development
School of Pharmacy
James K. Drennen, III
David A. Johnson
Alan W. Seadler
Carl A. Anderson
John D. Kirsch
Peter L. D. Wildfong
blend monitoring, chemometrics, multivariate calibration, near infrared, process analytical technology, tablet analysis
Interest in near-infrared (NIR) spectroscopy for pharmaceutical manufacturing has grown significantly in recent years. The prohibitive cost of method development, however, is an ongoing detriment to more routine use. Specifically, as shown in the preliminary chapters of this dissertation, the level of variation required for calibration exceeds that observed for most well-controlled pharmaceutical production processes. This insufficiency is often addressed by developing non-production samples to introduce leverage, but at high cost in labor and complexity. This dissertation presents two alternative methodologies for achieving cost-effective, efficient NIR method development: pure-component projection (PCP) and synthetic calibration. Both of the methodologies described in this dissertation were shown to be useful in reducing the resources required for quantitative NIR calibration.
The PCP method utilizes the information in the spectral characteristics of pure sample constituents, as well as knowledge of spectral interference patterns, to reduce NIR spectra to a univariate signal, thereby mitigating the need for non-production samples. Two alternative forms of PCP for NIR calibration are shown. The first method is based on multiplicative "shrinkage" of interference patters, and is referred to as generalized least-squares- (GLS-) PCP. The second method is based on additive correction of the error covariance structure based on net analyte signal (NAS) theory. Synthetic calibration, based on augmentation of pure-component and parallel-testing data with artificial interference spectra generated in silico, is introduced as a method to achieve efficient NIR calibration using off-the-shelf chemometric algorithms. Additionally, a method for estimating a calibration slope correction factor using only parallel test data is shown.
The theoretical and practical aspects of the PCP and synthetic calibration methodologies are described, and their performance is compared using both laboratory- and production-scale data from multi-component powder blending and tablet analysis applications. The results of this work demonstrate that, by using efficient calibration methods, accurate quantitative NIR calibration models for characterization of drug tablet quality can be created at significant cost savings, using only pure-component spectra and production-scale tablet samples.
Cogdill, R. (2007). Integration of Multivariate Simulation and Method Understanding for Efficient, Science-Based Process Analytical Systems Development (Doctoral dissertation, Duquesne University). Retrieved from https://dsc.duq.edu/etd/1704