A framework for enhancing process understanding using multivariate tools on commercial batch process data

Abstract

A lot of effort is made by pharmaceutical companies on the research and development of new pharmaceutical products and processes using the latest in quality by design tools, and process analytical technologies. Older pharmaceutical processes that were developed without the use of these tools are, however, somewhat neglected. Significant quantities of process data are routinely collected and stored but the information contained within this data is not extracted. Extensive literature on multivariate statistical process monitoring and control exists for exploring both batch and continuous process data. However, these methodologies rely on data from processes that are relatively well understood or controlled. Many industrial processes show batch to batch variability, which may be tolerated as it is not detrimental to the quality of the product, and the impact of this variability is not fully understood. The thesis presents a framework for exploring historical batch process data, to extract insights on where process control can be improved. The challenges presented with commercial process data are discussed. Multivariate tools such as multi-way principal component analysis are used to investigate variability in process data. The framework presented discusses the pre-processing steps necessary with batch process data, followed by outlier detection, and finally multivariate modelling of the data to identify where the process could benefit from improved understanding and control. This framework is demonstrated through the application to commercial process data from the active pharmaceutical drug substance manufacturing process of spironolactone at Piramal Healthcare, Morpeth, UK. In this case study, the process exhibits variability in drying times which traditional univariate data analysis has not been able to solve. The results demonstrated some of the challenges the use of the available data from commercial processes. Although the results from the multivariate data analysis did not show a significant statistical difference between the batches with long and short drying times, small differences were observed between these two groups. Further analysis of the crystallization process using infrared spectroscopic techniques which identified a potential root cause to the extended drying time.