Plenary lecture II - Real-time NIR monitoring of a pharmaceutical blending process through multivariate analysis-derived models

The Quality by Design (QbD) guideline of the USA Food & Drug Administration (FDA) and of the International Conference on Harmonisation (ICH) became lately the major driver of pharmaceutical processes optimization. The majority of these processes are complex and consequently multivariate. Although new insights have improved the knowledge on the phenomena taking place, it is not usually possible to develop deterministic models. Processes involving powders handling like the multi-component pharmaceutical formulations blending are common and the real-time monitoring of their physico-chemical attributes is challenging. This QbD initiative is nowadays possible through the use of Process Analytical Technologies (PAT). In this work we propose a multivariate analysis of a V-blender mixing unit operation using an in-line Near-Infra Red (NIR) measurement technique. For the NIR measurements, a system, consisting of an Axsun IntegraSpec XLP 410 spectrometer connected to an IP-65 encased optical measuring head (sampling probe) through a 1-meter length umbilical wire cord, was used. It uses the Diffuse Reflectance Sampling technology and provides a 40 nun spot size with a spectral range of 1350 rim to 1800 run. The methodology includes the following steps: (1) modification of a nominal 1 ft(3) (30 1) V-blender unit to accommodate Axsun's NIR spectroscopy system; (2) 3 experimental runs, each with different mixing time, while monitoring powder homogeneity with NIR spectroscopy; (3) acquisition of 10 powder samples after each run from predetermined locations in the V-blender, evaluated both with Axsun NIR spectrometer and current QA/QC Lab methods, to determine mixing end point and (4) data analysis using SIMPA-P+ and GRAMS chemometrics softwares. Two qualitative algorithms (Analysis of Spectral Variance and Distance Analysis using Hostelling T2) for real-time homogeneity determination are developed and their efficiency is evaluated. A quantitative model was derived and tested with success; it relies on the development of a Partial Least Squares (PLS) model in a principal component hyperspace which better describes the blending information. In all cases, the size of the acquired information is not comparable to the classical "thief analysis" and the result (prediction of the mixing end point) proved equally or more efficient than with actually employed quality control protocols. In addition, this information can be obtained in real-time using chemometric models. The time savings are huge when compared to classical laboratory analysis (i.e. High Pressure Liquid Chromatography). It is expected that any one of the presented NIR analyses can be beneficial on many aspects of pharmaceutical blending, such as: (1) Real-time quality monitoring of current manufacturing batches; (b) Improve process efficiency and performance by selecting adequate process parameters and blending time; (3) Quality by Design (QbD) initiatives during the development of blending processes for new formulas.