The use of statistics in understanding pharmaceutical manufacturing processes

Abstract

Industrial manufacturing processes for pharmaceutical products require a high level of understanding and control to demonstrate that the final product will be of the required quality to be taken by the patient. A large amount of data is typically collected throughout manufacture from sensors located around reaction vessels. This data has the potential to provide a significant amount of information about the variation inherent within the process and how it impacts on product quality. However to make use of the data, appropriate statistical methods are required to extract the information that is contained. Industrial process data presents a number of challenges, including large quantities, variable sampling rates, process noise and non-linear relationships. The aim of this thesis is to investigate, develop and apply statistical methodologies to data collected from the manufacture of active pharmaceutical ingredients (API), to increase the level of process and product understanding and to identify potential areas for improvement. Individual case studies are presented of investigations into API manufacture. The first considers prediction methods to estimate the drying times of a batch process using data collected early in the process. Good predictions were achieved by selecting a small number of variables as inputs, rather than data collected throughout the process. A further study considers the particle size distribution (PSD) of a product. Multivariate analysis techniques proved efficient at summarising the PSD data, to provide an understanding of the sources of variation and highlight the difference between two processing plants. Process capability indices (PCIs) are an informative tool to estimate the risk of a process failing a specification limit. PCIs are assessed and developed to be applied to data that does not follow a standard normal distribution. Calculating the capability from the percentiles of the data or the proportion of data outside of the specification limits has the potential to generate information about the capability of the process. Finally, the application of Bayesian statistical methods in pharmaceutical process development are investigated, including experimental design, process validation and process capability. A novel Bayesian method is developed to sequentially calculate the process capability when data is collected in blocks over time, thereby reducing the level of noise caused by small sample sizes.