Journal of pharmaceutical sciences

Designed blending for near infrared calibration.

PMID 25980978


Spectroscopic methods are increasingly used for monitoring pharmaceutical manufacturing unit operations that involve powder handling and processing. With that regard, chemometric models are required to interpret the obtained spectra. There are many ways to prepare artificial powder blend samples used in a chemometric model for predicting the chemical content. Basically, an infinite number of possible concentration levels exist in terms of the individual components. In our study, design of experiments for ternary mixtures was used to establish a suitable number of blend compositions that represents the entire mixture region of interest for a three component blend. Various experimental designs and their effect on the predictive power of a chemometric model for near infrared spectra were investigated. It was determined that a particular choice of experimental design could change the predictive power of a model, even with the same number of calibration experiments.