(11c) Application of Multivariate Statistical Process Control to a Drying Process for Continuous Pharmaceutical Manufacturing

1. Introduction

In the pharmaceutical industry, the method for the improvement of the production efficiency has been discussed and the documents on Quality by Design (QbD) and Process Analytical Technology (PAT) [1-4] were published by Food and Drug Administration (FDA) and International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH). Since then, online process monitoring and control technologies have attracted much attention. These technologies are important to realize the continuous manufacturing, and a new draft guideline that addresses quality aspects in the continuous manufacturing of medicinal products are published by the FDA in 2019 [5].

Data-driven multivariate analysis is an important part of PAT and many research papers were published [6, 7]. Latent variable method such as principal component analysis (PCA) and partial least squares (PLS) are popular, and they are applied to various kinds of processes such as crystallization, mixing, granulation, drying, and coating [6]. In this research, multivariate statistical process control (MSPC) is applied to the drying process. The research on MSPC in the drying process is rarely reported. As far as the authors know, only one paper [8] reported the application results of MSPC to a drying process and A.F. Silva et al [8] concluded that the abnormal conditions were successfully detected by MSPC. However, the rate of false positive is quite high: Q statistics are always over the threshold in some cases with normal operation conditions. This research aims to realize more accurate abnormal detection.

2. Experiments

Eight experiments were conducted under the normal operation conditions, and six experiments were conducted under the abnormal operation conditions. The abnormal operation conditions were caused by changing the flowrate, the temperature, and the humidity of the inlet air. Five variables, i.e. the inlet air flowrate, the inlet air temperature, the outlet air temperature, the product temperature, and the NIR spectrum of the product, were recorded every ten seconds.

3. PCA based MSPC

The data from seven experiments with normal operation conditions were used for model construction. Multiway PCA with one principal component was used since the drying process is batch process, and the model was constructed for every batch time. The maximum values of Hotteling’s T² and Q statistics for the model construction data are selected as the threshold for the abnormal detection. The model validation was conducted by using the data from one experiment with normal operation condition and six experiments with abnormal operation conditions. The MSPC method could classify the operation condition with the accuracy of more than 95%.

References

[1] ICH, ICH harmonised tripartite guideline - pharmaceutical development Q8 (R2) (2005)

[2] ICH, ICH harmonised tripartite guideline - quality risk management Q9 (2005)

[3] ICH, ICH harmonised tripartite guideline - pharmaceutical quality system Q10 (2008)

[4] FDA, Pharmaceutical cGMPs for the 21st century—a risk-based approach final report (2004)

[5] FDA, Quality considerations for continuous manufacturing guidance for Industry (2019)

[6] T. Rajalahti and O. M. Kvalheim “Multivariate data analysis in pharmaceutics: a tutorial review”, International journal of pharmaceutics, 417, (2011) 280-290

[7] G. M. Troup, C. Georgakis, “Process systems engineering tools in the pharmaceutical industry”, Computers and chemical engineering, 51 (2013) 157-171

[8] A.F. Silva et al. “Multivariate statistical process control of a continuous pharmaceutical twin-screw granulation and fluid bed drying process”, International journal of pharmaceutics, 528, (2017) 242-252