Multivariate curve resolution for control and modelling: an example of pharmaceutical blending process

Introduction
Blending processes are of utmost importance in the industrial production of pharmaceuticals or food commodities. As part of the multiple unit operations involved in the manufacturing process of pharmaceutical solid dosage forms, blending is critical in ensuring uniformity of composition in the final dosage form.
Controlling blend homogeneity is a necessary step in a drug product manufacturing process because problems incurred during blending can lead to inadequate tablet quality attributes. When a blending process is analysed, two main objectives can be of interest, namely: a) optimizing the process parameters to improve the blending operation and b) detecting when the blending process has reached a homogeneity standard set by external regulations or by the producer.
Multivariate Curve Resolution (MCR) is a chemometric method that has been widely used for process analysis in industrial, biological or chemical contexts. In this work, MCR is proposed for the double purpose of modelling and control the powder blending of a pharmaceutical formulation.

Experimental
Two different data sets have been used in this work. First, a three-component system (acetaminophen, microcrystalline cellulose and magnesium stearate) has been used to study the ability of MCR to model blending processes. Second, a five-component system (acetaminophen, lactose, microcrystalline cellulose, croscarmellose and magnesium stearate) has been investigated to assess the end-point detection of the blending process and, in general, to perform quantitative blending control. Ingredients were mixed simultaneously in a 3.5 quart, stainless steel, custom-made V-blender. A SpectralProbes Process NIR spectrometer was used to monitor the blending process in real-time and measurements were made through a sapphire window at the top of either arm of the blender. NIR spectra were formed by 100 absorption values between 1600 and 2400 nm in reflectance mode.

Results and Discussion
MCR has been applied to monitor pharmaceutical blending processes. Due to the flexibility in the analysis of augmented data sets (multiset analysis) and the application of constraints, blending modelling or control can be performed and work in on- and off- line modes. Classical MCR with natural non-negativity constraints can give information on the qualitative evolution of blending processes and provide information about the effects of modifying process parameters on the blending trajectory. When reference quantitative information about calibration blending runs is available, MCR analysis using a correlation constraint allows the determination of quantitative information of the blending profiles of unknown runs in real concentration units. This is useful to calculate homogeneity indicators and to detect the end-point of the blending process. Similar results are obtained whether working off- or on- line, which is very promising for the potential use of the MCR methodology for real in situ blending control.